Research map / 3D vision / Face and shape
Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging
Chongjie Ye, Yushuang Wu, Ziteng Lu, Jiahao Chang, Xiaoyang Guo, Jiaqing Zhou, Hao Zhao, Xiaoguang Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add DetailVerse synthetic data to image-normal training (ablation) | NiRNE · LUCES-MV | NE: 22.209 -> 21.837 | Improved |
| add DetailVerse synthetic data to image-normal training (ablation) | NiRNE · LUCES-MV | SNE: 28.324 -> 26.628 | Improved |
| add domain-specific training (real then synthetic fine-tuning of noisy stream) (ablation) | NiRNE · LUCES-MV | SNE: 29.690 -> 26.628 | Improved |
| add domain-specific training (ablation) | NiRNE · LUCES-MV | NE: 23.288 -> 21.837 | Improved |
| add dual-stream architecture (ablation) | NiRNE · LUCES-MV | SNE: 29.520 -> 26.628 | Improved |
| add dual-stream architecture (ablation) | NiRNE · LUCES-MV | NE: 21.918 -> 21.837 | No clear effect |
| add DS, DST, DV together (full vs w/o all) (ablation) | NiRNE · LUCES-MV | SNE: 35.997 -> 26.628 | Improved |
| use noise-injected regression NiRNE instead of GenPercept | NiRNE · LUCES-MV | NE: 28.050 -> 21.837 | Improved |
| use NiRNE instead of best diffusion method StableNormal | NiRNE · LUCES-MV | SNE: 37.045 -> 26.628 | Improved |
| bridge image-to-geometry with normal maps instead of direct Trellis generation (ablation) | Hi3DGen · 3D generation (qualitative) | visual fidelity | Improved |
| use sharper accurate normals instead of smoother or wrong normals from other estimators (ablation) | Hi3DGen · 3D generation (qualitative) | visual fidelity | Improved |
| add online normal regularization to latent diffusion (ablation) | NoRLD · 3D generation (qualitative) | visual fidelity | Improved |
Closest papers by meaning
- DIRECT-3D: Learning Direct Text-to-3D Generation on Massive Noisy 3D Data (CVPR 2024 · similarity 0.89)
- LaGeM: A Large Geometry Model for 3D Representation Learning and Diffusion (ICLR 2025 · similarity 0.89)
- Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image (NeurIPS 2024 · similarity 0.88)
- Cycle3D: High-quality and Consistent Image-to-3D Generation via Generation-Reconstruction Cycle (AAAI 2025 · similarity 0.88)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · similarity 0.88)
- HIFA: High-fidelity Text-to-3D Generation with Advanced Diffusion Guidance (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
